US6792442B1 - Signal processor and product-sum operating device for use therein with rounding function - Google Patents

Signal processor and product-sum operating device for use therein with rounding function Download PDF

Info

Publication number
US6792442B1
US6792442B1 US09/890,749 US89074901A US6792442B1 US 6792442 B1 US6792442 B1 US 6792442B1 US 89074901 A US89074901 A US 89074901A US 6792442 B1 US6792442 B1 US 6792442B1
Authority
US
United States
Prior art keywords
data
multiply
rounding
register
accumulate
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Fee Related
Application number
US09/890,749
Other languages
English (en)
Inventor
Takahiro Kumura
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
NEC Corp
Original Assignee
NEC Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by NEC Corp filed Critical NEC Corp
Assigned to NEC CORPORATION reassignment NEC CORPORATION ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: KUMURA, TAKAHIRO
Application granted granted Critical
Publication of US6792442B1 publication Critical patent/US6792442B1/en
Anticipated expiration legal-status Critical
Expired - Fee Related legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations

Definitions

  • the present invention relates to a signal processor and a multiply-accumulate unit with a rounding function for use in such a signal processor.
  • Signal processors read data from memory and process the read data in various ways, i.e., processes of addition, subtraction, logical operation, and multiplication.
  • the processing capability of signal processors are highly increased by incorporating a multiply-accumulate unit which can execute, in one processor cycle, multiply-accumulate operations that frequently appear in signal processing program such as image processing, sound processing, or the like.
  • FIG. 1 of the accompanying drawings shows a conventional signal processor having several execution units, registers, and memory.
  • the signal processor shown in FIG. 1 is introduced in “IEEE VLSI SIGNAL PROCESSING, VI”, pp. 93-101, 1993.
  • the conventional signal processor has eight 40-bit registers hereinafter referred to as “registers 50 ”), MAC (multiply-accumulate) unit 52 , MUX (multiplexer) 53 , ALU (arithmetic and logical unit) 54 , BSFT (barrel shift unit) 55 , X memory 57 x , and Y memory 57 y .
  • X memory 57 x and Y memory 57 y are hereinafter referred to as memory 57 x and memory 57 y , respectively.
  • Memory 57 x and memory 57 y are connected to registers 50 by respective data buses 58 x , 58 y .
  • MAC unit 52 , ALU 54 , MUX 53 , and BSFT 55 are connected to output lines 51 a , 51 b , and 51 c from registers 50 .
  • MAC unit 52 carries out multiply-accumulate operations.
  • ALU 54 carries out an arithmetic or logical operation using an immediate value imm. selected by MUX 53 or a value from registers 50 .
  • BSFT 55 carries out an arithmetic or logical shift using an immediate value imm. selected by MUX 53 or a value from registers 50 .
  • Multiply-accumulate operations that frequently appear in signal processing program are operations to perform a multiplication and an accumulation according to the following equation (1):
  • addend A on the right side of the equation (1) is the result of multiply-accumulate operations that are frequently performed, while it may be read from memory in some cases.
  • Operations that are represented by the equation (1) where the symbol “+” on the right side of the equation (1) is replaced with the symbol “ ⁇ ” are also referred to as multiply-accumulate operations.
  • multiplicand B and multiplier C on the right side of the equation (1) are usually expressed in 16-bit wide because of practical and economical reasons. Since the product of multiplicand B and multiplier C becomes 32-bit wide at maximum, each of addend A on the right side of the equation (1) and the sum A on the left side of the equation (1) need to be expressed in 32-bit wide or more.
  • general signal processors have 32-bit registers or more to save the results of multiply-accumulate operations.
  • two 16-bit data are held in one register of such a signal processor, they are placed in 15th-0th bits or 31st-16th bits of the register.
  • FIG. 2 shows a sequence to perform a multiply-accumulate operation with registers 50 and MAC unit 52 of the conventional signal processor shown in FIG. 1 .
  • Multiplicand B, multiplier C, and addend A on the right side of the equation (1) are read from memory connected to the signal processor into register 502 , register 503 , and register 501 , respectively.
  • Multiplicand B and multiplier C may be placed in either 31st-16th bits or 15th-0th bits of registers 502 , 503 . It is assumed here that multiplicand B is placed in 31st-16th bits of register 502 and multiplier C is placed in 15th-0th bits of register 503 . Addend A is placed in all the bits of register 501 . In FIG. 2, numerals shown beneath registers 501 , 502 , 503 indicate bit positions therein.
  • addend A is stored in ACC (accumulator) 523 of MAC unit 52 .
  • Multiplicand B and multiplier C are supplied to multiply unit 521 in MAC unit 52 , which calculates the product of multiplicand B and multiplier C.
  • the calculated product of multiplicand B and multiplier C is added to addend A from ACC 523 by adder/subtractor ( ⁇ ) 522 .
  • the sum produced by adder/subtractor 522 is temporarily stored in ACC 523 , and written back via output line 56 into register 501 which has stored addend A.
  • the above process occurs when the multiplicand or multiplier in a multiply-accumulate operation is used as the addend in another multiply-accumulate operation.
  • the result may possibly cause an overflow depending on the values of the addend, the multiplicand, and the multiplier.
  • the multiply-accumulate operation can be performed without an overflow if the addend, the multiplicand, and the multiplier are arranged in a suitable range.
  • multiplicand B, multiplier C, and addend A are read from memory connected to the signal processor into 31st-16th bits of register 502 , 15th-0th bits of register 503 , and 31st-16th bits of register 501 , respectively.
  • addend A expressed as fixed-point 16-bit data When addend A expressed as fixed-point 16-bit data is read into register 501 , the sign of addend A is inserted into 39th-32nd bits, addend A into 31st-16th bits, and “0” into 15th-0th bits.
  • a state in which the data are stored in registers 501 , 502 , 503 is referred to as state 50 n .
  • a state of the registers after the multiply-accumulate operation is referred to as state 50 n 1 .
  • state 50 n 2 A state of the registers after the result is rounded off is referred to as state 50 n 2 .
  • state 50 n 1 which follows state 50 n , the result A+B ⁇ C is stored in register 501 .
  • state 50 n 2 the result of the multiply-accumulate operation, which is 40-bit wide, is rounded off into 16 bits by ALU 54 , and the rounded result is stored in register 501 . Finally, the rounded result is stored in memory.
  • the first problem is that the data size of addend A read from memory and the data size of an addend required by MAC unit 52 are different from each other. Since addend A is 16-bit data, it has to be expanded into 40-bit data for multiply-accumulate operations. Therefore, two 16-bit addends cannot be placed in one register.
  • the second problem is that the data size of the result of the calculation performed by the MAC unit and the data size of the result when it is stored in memory are different from each other. Because the MAC unit of the conventional signal processor outputs a 40-bit result, when it is to be stored as 16-bit data into memory, the 40 bits need to be rounded off into 16 bits. Consequently, a rounding process has to be carried out in addition to the multiply-accumulate operation.
  • bus size between memory and the register is increased to 32 bits in order to improve performance of the conventional signal processor, then two 16-bit data can simultaneously be read through each data bus.
  • the conventional signal processor suffers from some problems with regard to the handling of 16-bit addends.
  • One of problems is that excess resources are occupied in multiply-accumulate operations where all inputs and outputs are 16-bit wide.
  • a 16-bit addend must be placed into a register with expanded to the width of the register in an appropriate manner in order to match the data size required by the MAC unit.
  • results of multiply-accumulate operations have the same data size as the size of registers, they need to be rounded off into 16-bit data in order to be stored into memory. This problem causes another problem in which the efficiency of data transfer between memory and the register cannot be increased.
  • An object of the present invention is to provide a signal processor which can solve the above problems and efficiently handle 16-bit addends, or to provide a multiply-accumulate unit with a rounding function for use in such a signal processor.
  • a more specific object of the present invention is to provide a signal processor which is able to execute efficiently 16-bit multiply-accumulate operations taking into account the position of an addend in a register, or to provide a multiply-accumulate unit with a rounding function for use in such a signal processor.
  • a signal processor based on the present invention includes a multiply-accumulate unit with a rounding function, which performs a multiply-accumulate operation on an addend, a multiplicand, and a multiplier.
  • the signal processor has a number of registers connected to the multiply-accumulate unit with the rounding function.
  • the multiply-accumulate unit with the rounding function comprises selecting inputting means for entering an addend supplied selectively from different positions in one of said registers, rounding means for performing a rounding process to convert data of a larger data size into data of a smaller data size on the result of the multiply-accumulate operation where the addend is selectively entered by said selecting inputting means, and selection outputting means for outputting the result of the multiply-accumulate operation rounded by said rounding means selectively to different positions in one of said registers.
  • a multiply-accumulate unit with a rounding function based on the present invention includes a multiply-accumulate unit for performing a multiply-accumulate operation on an addend, a multiplicand, and a multiplier.
  • the multiply-accumulate unit with the rounding function works together with a number of registers connected to the unit, and comprises selecting inputting means for entering an addend supplied selectively from different positions in one of said registers which is connected externally thereto, rounding means for performing a rounding process to convert data of a larger data size into data of a smaller data size on the result of the multiply-accumulate operation based on the addend selectively entered by said selecting inputting means, and selection outputting means for outputting the result of the multiply-accumulate operation rounded by said rounding means selectively to different positions in one of said registers.
  • the multiply-accumulate unit with the rounding function based on the present invention has a selection inputting and expanding means, a rounding and selection outputting means, and a multiply-accumulate unit.
  • the multiply-accumulate unit with the rounding function has its operation mode which is controlled by two kinds of signals Round, Position. If control signal Round is “0”, then the multiply-accumulate unit with the rounding function based on the present invention operates as the conventional multiply-accumulate unit.
  • control signal Round is “1”
  • the multiply-accumulate unit with the rounding function based on the present invention operates differently depending on control signal Position.
  • the selection inputting and expanding means expands an addend into a data which has the data size required by the multiply-accumulate unit based on the control signal Position which indicates the position of the addend represented by 16-bit wide data in the externally connected register.
  • the rounding and selection outputting means rounds off the result of the multiply-accumulate operation whose data size corresponds to the data size of the register into 16-bit data, and then outputs the rounded data to the position of the addend in the register which is indicated by control signal Position.
  • FIG. 1 is a block diagram of a conventional signal processor
  • FIG. 2 is a block diagram showing a sequence of performing a multiply-accumulate operation on an addend of 40-bit data, a multiplicand and a multiplier, each of 16-bit data, with the conventional signal processor;
  • FIG. 3 is a block diagram showing registers states in a sequence of performing a multiply-accumulate operation on an addend, a multiplicand, and a multiplier, each of 16-bit data, with the conventional signal processor;
  • FIG. 4 is a block diagram of an arrangement based on an embodiment of the present invention.
  • FIG. 5 is a diagram showing the manner in which a selection inputting and expanding means in FIG. 4 operates;
  • FIG. 6 is a diagram showing the manner in which a rounding and selection outputting means in FIG. 4 operates;
  • FIG. 7 is a block diagram of a multiply-accumulate unit with a rounding function based on an embodiment of the present invention.
  • FIG. 8 is a block diagram showing registers states in a sequence of performing a multiply-accumulate operation on an addend, a multiplicand, and a multiplier, each of 16-bit data, with the multiply-accumulate unit with a rounding function shown in FIG. 7 .
  • FIG. 4 is a block diagram showing an arrangement based on an embodiment of the present invention.
  • Rounding-MAC unit 4 is arranged so as to be able to efficiently handle 16-bit addends. More specifically, Rounding-MAC unit 4 is arranged so as to be able to perform multiply-accumulate operations on each addend in register 1 without affecting each other, where the register is 32-bit wide or more and two separate 16-bit addends are placed in the register.
  • Rounding-MAC unit 4 operates with registers connected thereto, and reads three data, i.e., an addend, a multiplicand, and a multiplier from the registers. Rounding-MAC unit 4 has three outputs as the results of operations. Operations performed by Rounding-MAC unit 4 are controlled by two kinds of signals Round, Position.
  • An addend is stored in either 31st-16th bits or 15th-0th bits of 40-bit register 1 that is connected to Rounding-MAC unit 4 .
  • a multiplicand and a multiplier are stored in other registers, and entered into Rounding-MAC unit 4 .
  • FIG. 4 shows 40-bit register 1 which stores an addend, but does not show registers which store the a multiplicand and a multiplier.
  • Data stored in 15th-0th bits of 40-bit register 1 , data stored in 31st-16th bits thereof, and data stored in 39th-32nd bits thereof are entered respectively as input data 46 L, input data 46 H, and input data 46 E into MAC unit 4 .
  • the multiplicand is entered as 16-bit data 45 into MAC unit 4
  • the multiplier is entered as 16-bit data 44 into MAC unit 4 .
  • Data 46 L stored in 15th-0th bits of 40-bit register 1 , data 46 H stored in 31st-16th bits thereof, and data 46 E stored in 39th-32nd bits thereof are entered as addends or parts of an addend into Rounding-MAC unit 4 .
  • Multiplicand 45 and multiplier 44 are entered into Rounding-MAC unit 4 .
  • Outputs 47 L, 47 H, 47 E are written back as outputs 47 L, 47 H, 47 E, respectively, into 40-bit register 1 .
  • Outputs 47 L, 47 H are represented as 16-bit data
  • output 47 E is represented as 8-bit data.
  • Output 47 L is output to 15th-0th bits of 40-bit register 1 .
  • Output 47 H is output to 31st-16th bits of 40-bit register 1 .
  • Output 47 E is output to 39th-32nd bits of 40-bit register 1 .
  • Control signal Position is a signal to indicate the position of an addend in 40-bit register 1 . If a desired addend is placed in 31st-16th bits of 40-bit register 1 , then control signal Position should be set to “1”, and if a desired addend is placed in 15th-0th bits of 40-bit register 1 , then control signal Position should be set to “0”.
  • Control signal Round is a signal to indicate whether Rounding-MAC unit 4 is to perform a rounding process or not.
  • the rounding process is a process for converting data of a larger data size to data of a smaller data size (e.g. 32-bit data to 16-bit data).
  • a reverse process i.e., a process for converting data of a smaller data size to data of a larger data size, is referred to as an expanding process. If the rounding process is to be carried out, then control signal Round is set to “1”, and the rounding process is not to be carried out, then control signal Round is set to “0”.
  • Rounding-MAC unit 4 comprises MAC unit 41 , selection inputting and expanding means 42 , and rounding and selection outputting means 43 .
  • Selection inputting and expanding means 42 selects data from input data 46 E, 46 H, 46 L in register 1 , expands the selected data, and outputs 40-bit data. Selection inputting and expanding means 42 is controlled by control signals Position, Round. MAC unit 41 performs a multiply-accumulate operation on multiplier 44 , multiplicand 45 , and addend produced by selection inputting and expanding means 42 .
  • Rounding and selection outputting means 43 rounds off an output, which is generated by MAC unit 41 , from 40 bits into 16 bits, and outputs the 16-bit data to a designated position in 40-bit register 1 .
  • Rounding and selection outputting means 43 is controlled by control signals Position, Round.
  • Rounding-MAC unit 4 based on the present embodiment is a combination of MAC unit 41 , selection inputting and expanding means 42 , and rounding and selection outputting means 43 for improving the handling of 16-bit data.
  • Selection inputting and expanding means 42 produces 40-bit data 48 based on input data 46 E, 46 H, 46 L entered from 40-bit register 1 under the control of control signals Round, Position.
  • control signal Round is “0”, no rounding process is performed.
  • Rounding-MAC unit 4 has to behave as if it was the conventional MAC unit.
  • selection inputting and expanding means 42 outputs input data 46 E, 46 H, 46 L entered from 40-bit register 1 as one 40-bit data, unchanged, to MAC unit 41 .
  • control signal Round is “1”
  • Rounding-MAC unit 4 operates differently depending on control signal Position. If control signal Round is “1” and control signal Position is “0”, then selection inputting and expanding means 42 works on the assumption that the addend is placed in 15th-0th bits of 40-bit register 1 . In this case, selection inputting and expanding means 42 expands the addend 46 L from 15th-0th bits of 40-bit register 1 into 40-bit data and outputs the 40-bit data to MAC unit 41 .
  • control signal Round is “1” and control signal Position is “1”
  • Rounding-MAC unit 4 selection inputting and expanding means 42 works on the assumption that the addend is placed in the 31st-16th bits of 40-bit register 1 .
  • selection inputting and expanding means 42 expands the addend 46 H from 31st-16th bits of 40-bit register 1 into 40-bit data and outputs the 40-bit data to MAC unit 41 .
  • MAC unit 41 calculates a product of 16-bit data 44 , 45 , adds the product to 40-bit data 48 , and outputs the sum as 40-bit data 49 .
  • rounding and selection outputting means 43 rounds off 40-bit data 49 representing the result of the multiply-accumulate operation performed by MAC unit 41 , and outputs the rounded result as 16-bit data to 40-bit register 1 .
  • Rounding and selection outputting means 43 determines whether to round off 40-bit data 49 or not, based on control signal Round. Rounding and selection outputting means 43 determines which bit position of 40-bit register 1 to output the rounded 16-bit data to, based on control signal Position. If control signal Round is “0”, then rounding and selection outputting means 43 does not round off 40-bit data 49 , but divides 40-bit data 49 into outputs 47 E, 47 H, 47 L and output them to 40-bit register 1 .
  • rounding and selection outputting means 43 rounds off 40-bit data 49 into 16-bit data, and outputs the 16-bit data as output 47 L to 15th-0th bits of 40-bit register 1 .
  • outputs 47 H, 47 E are not operated upon, the data in 39th-16th bits of 40-bit register 1 is not changed.
  • control signal Round is “1” and control signal Position is “1”
  • rounding and selection outputting means 43 rounds 40-bit data 49 into 16-bit data, and outputs the 16-bit data as output 47 H to 31st-16th bits of 40-bit register 1 .
  • outputs 47 L, 47 E are not operated upon, the data in 39th-32nd bits and 15th-0th bits of 40-bit register 1 are not changed.
  • FIG. 5 shows the manner in which selection inputting and expanding means 42 shown in FIG. 4 operates. Let us describe behavior of selection inputting and expanding means 42 with reference to FIG. 5 .
  • Selection inputting and expanding means 42 serves the purpose of generating 40-bit data from 40-bit register 1 in view of whether data should be rounded off or not and the position in 40-bit register 1 of the added required in the multiply-accumulate operation.
  • control signal Round Whether data should be rounded off or not by selection inputting and expanding means 42 is determined by control signal Round, and the position in 40-bit register 1 of the added is determined by control signal Position.
  • selection inputting and expanding means 42 produces 40-bit data 48 from input data 46 E, 46 H, 46 L entered from 40-bit register 1 and outputs 40-bit data 48 regardless of control signal Position.
  • 16-bit data 46 L is represented as 15th-0th bits of 40-bit data 48
  • 16-bit data 46 H as 31st-16th bits of 40-bit data 48
  • 8-bit data 46 E as 39th-32nd bits of 40-bit data 48 , respectively.
  • control signal Round is “1”
  • selection inputting and expanding means 42 operates differently depending on control signal Position. If control signal Round is “1” and control signal Position is “0”, the addend is placed in 15th-0th bits of 40-bit register 1 . Selection inputting and expanding means 42 expands 16-bit data 46 L from 15th-0th bits of 40-bit register 1 into 40-bit data 48 .
  • control signal Round is “1” and control signal Position is “1”
  • the addend is placed in 31st-16th bits of 40-bit register 1 .
  • Selection inputting and expanding means 42 expands 16-bit data 46 H from 31st-16th bits of 40-bit register 1 into 40-bit data 48 .
  • selection inputting and expanding means 42 expands 16-bit data 46 L or 16-bit data 46 H into 40-bit data 48 as follows: First, selection inputting and expanding means 42 places 16-bit data to be expanded into 31st-16th bits of 40-bit data 48 .
  • selection inputting and expanding means 42 extracts a sign bit of 16-bit data to be expanded, and places the sign bit into 39th-32nd bits of 40-bit data 48 . Because the sign bit is usually represented as 1-bit information, the sign bit is repeatedly inserted into 39th-32nd bits of 40-bit data 48 . Finally, selection inputting and expanding means 42 inserts “0” into 15th-0th bits of 40-bit data 48 .
  • FIG. 6 shows the manner in which rounding and selection outputting means 43 shown in FIG. 4 operates. Let us describe behavior of rounding and selection outputting means 43 with reference to FIG. 6 . Rounding and selection outputting means 43 is supplied with the result 49 of the multiply-accumulate operation performed by MAC unit 41 .
  • Rounding and selection outputting means 43 serves the purpose of outputting entered 40-bit data 49 to 40-bit register 1 in view of whether data should be rounded off or not and the position in 40-bit register 1 of the added used in the multiply-accumulate operation.
  • Whether data should be rounded off or not by rounding and selection outputting means 43 is determined by control signal Round, and the position in 40-bit register 1 of the added is determined by control signal Position.
  • control signal Round is “0”
  • rounding and selection outputting means 43 does not round off 40-bit data 49 , but outputs 40-bit data 49 , unchanged, to 40-bit register 1 .
  • output 16-bit data 47 L is represented as 15th-0th bits of 40-bit data 49 , 16-bit data 47 H as 31st-16th bits of 40-bit data 49 , and output 8-bit data 47 E as 39th-32nd bits of 40-bit data 49 .
  • rounding and selection outputting means 43 operates differently depending on control signal Position. If control signal Round is “1” and control signal Position is “0”, the addend is placed in 15th-0th bits of 40-bit register 1 . Rounding and selection outputting means 43 rounds off 40-bit data 49 into 16-bit data, and outputs the 16-bit data as output 47 L to the 15th-0th bits of 40-bit register 1 . At this time, outputs 47 H, 47 E are not output.
  • control signal Round is “1” and control signal Position is “1”
  • the addend is placed in 31st-16th bits of 40-bit register 1 .
  • Rounding and selection outputting means 43 rounds off input 40-bit data 49 into 16-bit data, and outputs the 16-bit data as output 47 H to 31st-16th bits of 40-bit register 1 . At this time, outputs 47 L, 47 E are not output.
  • the present embodiment may be modified to change the register width and the data size for allowing M-bit registers to handle N-bit data where N and M are integer numbers and N ⁇ M/2.
  • register 1 comprises M-bit register
  • each of data 2 , 4 , 44 , 45 , 46 H, 46 L, 47 H, 47 L comprises N-bit data
  • each of data 46 E, 47 E comprises (M ⁇ 2N)-bit data
  • each of data 48 , 49 comprises M-bit data.
  • MAC unit 41 calculates a product of N-bit data 44 , 45 , adds the product to M-bit data 48 received from selection inputting and expanding means 42 .
  • Sum 49 calculated by MAC unit 41 is rounded off by rounding and selection outputting means 43 , which writes N-bit data back into register 1 .
  • FIG. 7 shows a block diagram showing the multiply-accumulate unit with the rounding function based on the present embodiment and three registers connected to the unit.
  • 40-bit registers 61 , 62 , 63 are connected to Rounding-MAC unit 4 , and 31st-16th bits or 15th-0th bits of registers 62 , 63 are selected by MUXs (multiplexers) 64 , 65 and entered to Rounding-MAC unit 4 .
  • registers 61 , 62 , 63 are arranged to read data from memory (not shown) and save data in memory.
  • FIG. 8 shows register states in a sequence of performing a multiply-accumulate operation on an addend, a multiplicand, and a multiplier, each of 16-bit data, with Rounding-MAC unit 4 shown in FIG. 7 . Let us describe behavior of Rounding-MAC unit 4 with reference to FIGS. 7 and 8.
  • Rounding-MAC unit 4 reads addend A from register 61 , performs a multiply-accumulate operation on added A, multiplicand B selected from register 62 by MUX 64 , and multiplier C selected from register 63 by MUX 65 , and rounds off the result of the multiply-accumulate operation.
  • the rounded result is written back into register 61 .
  • the result is written in 31st-16th bits of register 61 , whereas data in 15th-0th bits of register 61 remains unchanged. This state is referred to as state 6 n 2 in FIG. 8 .
  • Register 61 in state 6 n 2 stores, respectively in 31st-16th bits and 15th-0th bits thereof, the results of these operations A+B ⁇ C, D+E ⁇ F as they have been rounded off into 16-bit data. These data can directly be stored in memory.
  • register 61 is fixedly combined with data 46 L, 46 H, 46 E, 47 E, 47 H, 47 L, register 62 with multiplicand 16-bit data 45 via MUX 64 , and register 63 with multiplier 16-bit data 44 via MUX 65 .
  • the overall arrangement may be modified to freely select these combinations.
  • Such a modification is particularly effective for using Rounding-MAC unit 4 in a signal processor having a plurality of registers. It is also possible to use Rounding-MAC unit 4 as a circuit that is not included in a signal processor.
  • Rounding-MAC unit 4 makes it possible to perform a multiply-accumulate operation on a 16-bit addend placed in 31st-16th bits or 15th-0th bits of a register, without affecting other bits, as shown in FIG. 8 . This allows different 16-bit addends to be placed together in the register, resulting in reduction of the number of registers used for multiply-accumulate operations.
  • Rounding-MAC unit 4 is incorporated in a signal processor having 32-bit registers or more, then it is possible for Rounding-MAC unit 4 to perform multiply-accumulate operations properly even when it simultaneously reads two 16-bit addends from memory and places them together in one register. Consequently, the 32-bit data transfer capability for data transfer between memory and registers can be used to its full advantage.
  • the output y i is calculated and coefficients w j are updated according to the following equations:
  • ⁇ represents a very small positive constant
  • j 0, 1, . . . , T ⁇ 1.
  • a basic process required for updating each of the filter coefficients represented by the equation (4) comprises the following four steps of:
  • each of x i ⁇ j , w j , e i is usually represented as 16-bit data
  • y i is usually represented as 32-bit data or more.
  • the data size is required by specifications of a multiply-accumulate unit incorporated in signal processors.
  • the multiply-accumulate operation performed in the step (3) is an operation where all the data are 16-bit wide.
  • the 32-bit data transfer capability is used to read w j in the step (2), then two 16-bit data, w j , w j+1 are placed in 31st-16th bits and 15th-0th bits of one register, respectively.
  • a processor with the conventional multiply-accumulate unit is unable to perform a multiply-accumulate operation on w j , w j in one register thus placed as it is.
  • the signal processor which incorporates Rounding-MAC unit 4 based on the present embodiment can immediately perform the multiply-accumulate operation on w j , w j ⁇ 1 thus stored.
  • the 32-bit data transfer capability can be used to save w j , w j+1 in memory in the step (4) immediately after the multiply-accumulate operation.
  • the signal processor having 32-bit data buses performs a multiply-accumulate operation on 16-bit data
  • its 32-bit data transfer capability to transfer data between memory and registers can be used to its full advantage due to Rounding-MAC unit 4 .

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Computational Mathematics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Algebra (AREA)
  • Complex Calculations (AREA)
US09/890,749 1999-02-03 2000-02-03 Signal processor and product-sum operating device for use therein with rounding function Expired - Fee Related US6792442B1 (en)

Applications Claiming Priority (3)

Application Number Priority Date Filing Date Title
JP02567499A JP3336986B2 (ja) 1999-02-03 1999-02-03 信号処理プロセッサ及びそれに用いる丸め機能付き積和演算器
JP11-025674 1999-02-03
PCT/JP2000/000584 WO2000046692A1 (fr) 1999-02-03 2000-02-03 Processeur de signaux et additionneur-multiplicateur a fonction d'arrondi utilise dans ce dernier

Publications (1)

Publication Number Publication Date
US6792442B1 true US6792442B1 (en) 2004-09-14

Family

ID=12172340

Family Applications (1)

Application Number Title Priority Date Filing Date
US09/890,749 Expired - Fee Related US6792442B1 (en) 1999-02-03 2000-02-03 Signal processor and product-sum operating device for use therein with rounding function

Country Status (6)

Country Link
US (1) US6792442B1 (ja)
EP (1) EP1197874B1 (ja)
JP (1) JP3336986B2 (ja)
CA (1) CA2361451A1 (ja)
DE (1) DE60035999T2 (ja)
WO (1) WO2000046692A1 (ja)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060179166A1 (en) * 2005-02-01 2006-08-10 Tomoo Nagai Operation apparatus

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP3984790B2 (ja) * 2001-01-15 2007-10-03 日本電気株式会社 ビタビ復号処理装置

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS59197936A (ja) 1983-04-25 1984-11-09 Fujitsu Ltd デイジタル信号処理方式
US4876660A (en) * 1987-03-20 1989-10-24 Bipolar Integrated Technology, Inc. Fixed-point multiplier-accumulator architecture
JPH0520352A (ja) 1991-07-12 1993-01-29 Nippon Steel Corp 加算回路
US5363322A (en) * 1991-04-02 1994-11-08 Motorola, Inc. Data processor with an integer multiplication function on a fractional multiplier
JPH0756712A (ja) 1992-07-03 1995-03-03 Sgs Thomson Microelectron Sa 数値計算を実行するプロセスおよびそのプロセスを実現する演算装置
JPH08137669A (ja) 1994-11-04 1996-05-31 Fujitsu Ltd 演算回路
JPH09274612A (ja) 1996-04-05 1997-10-21 Hitachi Ltd ベクトル処理装置
JPH10134032A (ja) 1996-09-04 1998-05-22 Seiko Epson Corp 情報処理回路、マイクロコンピュータ及び電子機器
US5862067A (en) * 1995-12-29 1999-01-19 Intel Corporation Method and apparatus for providing high numerical accuracy with packed multiply-add or multiply-subtract operations

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101794213B (zh) * 1995-08-31 2014-09-17 英特尔公司 控制移位分组数据的位校正的装置
GB2317465B (en) * 1996-09-23 2000-11-15 Advanced Risc Mach Ltd Data processing apparatus registers.

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS59197936A (ja) 1983-04-25 1984-11-09 Fujitsu Ltd デイジタル信号処理方式
US4876660A (en) * 1987-03-20 1989-10-24 Bipolar Integrated Technology, Inc. Fixed-point multiplier-accumulator architecture
US5363322A (en) * 1991-04-02 1994-11-08 Motorola, Inc. Data processor with an integer multiplication function on a fractional multiplier
JPH0520352A (ja) 1991-07-12 1993-01-29 Nippon Steel Corp 加算回路
JPH0756712A (ja) 1992-07-03 1995-03-03 Sgs Thomson Microelectron Sa 数値計算を実行するプロセスおよびそのプロセスを実現する演算装置
JPH08137669A (ja) 1994-11-04 1996-05-31 Fujitsu Ltd 演算回路
US5862067A (en) * 1995-12-29 1999-01-19 Intel Corporation Method and apparatus for providing high numerical accuracy with packed multiply-add or multiply-subtract operations
JPH09274612A (ja) 1996-04-05 1997-10-21 Hitachi Ltd ベクトル処理装置
JPH10134032A (ja) 1996-09-04 1998-05-22 Seiko Epson Corp 情報処理回路、マイクロコンピュータ及び電子機器

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060179166A1 (en) * 2005-02-01 2006-08-10 Tomoo Nagai Operation apparatus
US7673117B2 (en) * 2005-02-01 2010-03-02 Sony Corporation Operation apparatus

Also Published As

Publication number Publication date
DE60035999T2 (de) 2007-12-20
EP1197874A4 (en) 2005-02-09
JP3336986B2 (ja) 2002-10-21
EP1197874A1 (en) 2002-04-17
JP2000222174A (ja) 2000-08-11
DE60035999D1 (de) 2007-09-27
CA2361451A1 (en) 2000-08-10
WO2000046692A1 (fr) 2000-08-10
EP1197874B1 (en) 2007-08-15

Similar Documents

Publication Publication Date Title
US6009451A (en) Method for generating barrel shifter result flags directly from input data
EP0754998B1 (en) An arithmetic unit
US5991785A (en) Determining an extremum value and its index in an array using a dual-accumulation processor
US5844830A (en) Executing computer instrucrions by circuits having different latencies
EP0517429A2 (en) CPU with integrated multiply/accumulate unit
US20030041082A1 (en) Floating point multiplier/accumulator with reduced latency and method thereof
US5892699A (en) Method and apparatus for optimizing dependent operand flow within a multiplier using recoding logic
JPH06236257A (ja) データ・プロセッサにおいて浮動小数点実行ユニット内で後正規化を実行する方法および装置
US5623683A (en) Two stage binary multiplier
US6941334B2 (en) Higher precision divide and square root approximations
JPH08123769A (ja) 並列プロセッサ
JPS63123125A (ja) 浮動小数点数の加算装置
US6792442B1 (en) Signal processor and product-sum operating device for use therein with rounding function
EP1936492A1 (en) SIMD processor with reduction unit
JP2003241960A (ja) プロセッサ、該プロセッサにおける演算命令処理方法及び演算制御方法
JP3579087B2 (ja) 演算器およびマイクロプロセッサ
US5539684A (en) Method and apparatus for calculating floating point exponent values
US20030233384A1 (en) Arithmetic apparatus for performing high speed multiplication and addition operations
JPH04172526A (ja) 浮動小数点除算器
WO2008077803A1 (en) Simd processor with reduction unit
US7237000B2 (en) Speed of execution of a conditional subtract instruction and increasing the range of operands over which the instruction would be performed correctly
JPH05324694A (ja) 再構成可能並列プロセッサ
JP3441847B2 (ja) データメモリを有するプロセッサ
JP3723311B2 (ja) 並列演算プロセッサ
Fayyazi et al. An Efficient CPU Architecture for DSP Processor

Legal Events

Date Code Title Description
AS Assignment

Owner name: NEC CORPORATION, JAPAN

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:KUMURA, TAKAHIRO;REEL/FRAME:012142/0173

Effective date: 20010724

FEPP Fee payment procedure

Free format text: PAYOR NUMBER ASSIGNED (ORIGINAL EVENT CODE: ASPN); ENTITY STATUS OF PATENT OWNER: LARGE ENTITY

CC Certificate of correction
FPAY Fee payment

Year of fee payment: 4

FPAY Fee payment

Year of fee payment: 8

REMI Maintenance fee reminder mailed
LAPS Lapse for failure to pay maintenance fees
STCH Information on status: patent discontinuation

Free format text: PATENT EXPIRED DUE TO NONPAYMENT OF MAINTENANCE FEES UNDER 37 CFR 1.362

FP Lapsed due to failure to pay maintenance fee

Effective date: 20160914